how to train with my own dataset?

[Cabbagehust2507]

I have a dataset including many folders, each folder (as a class) contains images. So I want to train with my own dataset but I don't know how to set up my data structure. Thank you so much!

[hysts]

Hi, @Cabbagehust2507

I think you can use torchvision.datasets.ImageFolder. You need to modify this file to make your dataset available in this repository. See this part.

If your directory is arranged in exactly the same way as ImageNet, which looks like the following, you can use config files for ImageNet like this without modifying any code. In that case, you need to specify dataset.dataset_dir, dataset.image_size, dataset.n_classes, etc. in the config file, and you need to keep dataset.name as ImageNet.

.
├── train
│   ├── n01440764
│   │   ├── n01440764_10026.JPEG
│   │   ├── n01440764_10027.JPEG
│   │   ├── n01440764_10029.JPEG
...
│   │   └── n01440764_9981.JPEG
│   ├── n01443537
│   │   ├── n01443537_10007.JPEG
│   │   ├── n01443537_10014.JPEG
│   │   ├── n01443537_10025.JPEG

...
│       ├── n15075141_9993.JPEG
│       └── n15075141_999.JPEG
└── val
    ├── n01440764
    │   ├── ILSVRC2012_val_00000293.JPEG
    │   ├── ILSVRC2012_val_00002138.JPEG
    │   ├── ILSVRC2012_val_00003014.JPEG
...

FYI, here is a great repository for image classification. You might want to take a look at it.

[Cabbagehust2507]

@hysts Thank you so much!